A Video Quality Evaluation Method Based on Transformers

This article proposes a dual channel video quality evaluation method based on Transformers. The jitter, exposure, and compression during video capture and transmission can all affect the quality of the video. In response to the spatial effects of exposure and defocus, a network based on Swin Transformers is used to extract multi-scale spatial features. The Transformer network is used to extract spatial and temporal features from 2D and 3D perspectives, respectively. In response to factors such as jitter, a backbone network based on video spin transformers is used to extract motion information. At the same time, considering the human visual system, an improved version of the time effect module is proposed, providing a new time memory attention that can more effectively map video frame quality scores. The model proposed in this article has been tested on multiple datasets, and the results show that the method has good performance.